Cancellable Loop

A minimal example that shows how to write a skill script that honours notifications/cancelled from the MCP client.

The pattern is simple: call check_cancelled() at the top of every iteration of a long-running loop. When the dispatcher installs a CancelToken and the client cancels the request, check_cancelled() raises CancelledError and the script unwinds cleanly. Outside of a request context (REPL, unit tests) check_cancelled() is a no-op, so the same script remains easy to run in isolation.

Tools

• cancellable_loop__count — Iterate iterations times, sleeping sleep_ms milliseconds per step, checking for cancellation each iteration.

Example

{"name": "cancellable_loop__count", "arguments": {"iterations": 100, "sleep_ms": 50}}
# → {"success": true, "message": "Completed 100 iterations", "context": {"iterations": 100}}

If the client sends notifications/cancelled while the loop is running, the next check_cancelled() call raises CancelledError and the @skill_entry wrapper converts it into a standard error dict.

Related

• dcc_mcp_core.check_cancelled — the API this skill demonstrates.
• Issue #329 — cooperative cancellation checkpoints.
• Issue #318 — async dispatcher integration (wires the CancelToken).